Setup Hooks

Configure Claude Code hooks for your project. Hooks are deterministic automation that runs before or after Claude's tool calls.

Usage

/setup-hooks

Background

Claude Code supports these hook events:

• PreToolUse: Runs BEFORE a tool call. Can block the call.
• PostToolUse: Runs AFTER a tool call completes.
• Notification: Runs when Claude sends a notification.
• Stop: Runs when Claude finishes its turn.

Hooks are configured in .claude/settings.json under the hooks key.

Procedure

Step 1: Present Hook Options

Show the user available hook patterns:

Which hooks would you like to set up?

1. Auto-format after edits — Runs your formatter after every file edit
2. Block dangerous commands — Prevents rm -rf, git push --force, etc.
3. Lint on save — Runs your linter after every file write
4. Custom hook — Define your own

Select one or more (e.g., "1,2"):

Step 2: Configure Selected Hooks

Hook 1: Auto-Format After Edits

1. Detect the project's formatter (see detection logic below)
2. Create a PostToolUse hook for the Edit and Write tools:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "command": "[formatter-command] $CLAUDE_FILE_PATH",
        "timeout": 10000
      }
    ]
  }
}

Formatter detection:

• Node.js: Check for prettier → npx prettier --write
• Node.js: Check for biome → npx biome format --write
• Python: Check for ruff → ruff format
• Python: Check for black → black
• Rust: cargo fmt --
• Go: gofmt -w

Hook 2: Block Dangerous Commands

Create a PreToolUse hook for the Bash tool:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "command": "node .claude/hooks/block-dangerous.js",
        "timeout": 5000
      }
    ]
  }
}

Create the blocking script at .claude/hooks/block-dangerous.js:

// Block dangerous commands in Claude Code
// Reads tool input from stdin, exits non-zero to block

const fs = require('fs');
const input = JSON.parse(fs.readFileSync('/dev/stdin', 'utf8'));
const cmd = input.tool_input?.command || '';

const BLOCKED_PATTERNS = [
  /rm\s+-rf\s+[\/~]/,           // rm -rf with absolute or home path
  /git\s+push\s+.*--force/,     // force push
  /git\s+reset\s+--hard/,       // hard reset
  /drop\s+table/i,              // SQL drop table
  /drop\s+database/i,           // SQL drop database
  />\s*\/dev\/sd/,              // writing to block devices
  /mkfs\./,                     // formatting filesystems
  /dd\s+if=/,                   // disk destroyer
];

for (const pattern of BLOCKED_PATTERNS) {
  if (pattern.test(cmd)) {
    console.error(`BLOCKED: Command matches dangerous pattern: ${pattern}`);
    process.exit(1);
  }
}

process.exit(0);

Note for Windows: The script above uses Unix conventions. On Windows, adjust stdin reading:

const input = JSON.parse(require('fs').readFileSync(0, 'utf8'));

Hook 3: Lint on Save

Similar to auto-format, but runs the linter:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write",
        "command": "[lint-command] $CLAUDE_FILE_PATH",
        "timeout": 15000
      }
    ]
  }
}

Step 3: Install Hooks

1. Read the current .claude/settings.json
2. Merge the new hooks into the existing configuration (don't overwrite existing hooks)
3. Write the updated settings file
4. If the dangerous-command hook was selected, create the .claude/hooks/ directory and script

Step 4: Verify

Tell the user the hooks are installed and how to test them:

Hooks installed successfully!

To verify:
- Auto-format: Edit any file and check it gets formatted
- Block dangerous: Try running `rm -rf /` (it will be blocked)
- Lint on save: Write a file and check for lint output

To view all hooks: /hooks
To disable: Remove the hook entry from .claude/settings.json